kobart-transformers 0.1.3

Kobart model on huggingface transformers

KoBart-Transformers

• SKT에서 공개한 KoBart를 편리하게 사용할 수 있게 transformers로 포팅하였습니다.
  
  

Install

pip install kobart-transformers

Tokenizer

• PreTrainedTokenizerFast를 이용하여 구현되었습니다.

>>> from kobart_transformers import get_kobart_tokenizer
>>> kobart_tokenizer = get_kobart_tokenizer()
>>> kobart_tokenizer.tokenize("안녕하세요. 한국어 BART 입니다.🤣:)l^o")
['▁안녕하', '세요.', '▁한국어', '▁B', 'A', 'R', 'T', '▁입', '니다.', '🤣', ':)', 'l^o']

Model

• BartModel을 이용하여 구현되었습니다.
• BartModel.from_pretrained("hyunwoongko/kobart")와 동일합니다.

>>> from kobart_transformers import get_kobart_model, get_kobart_tokenizer
>>> # from transformers import BartModel

>>> kobart_tokenizer = get_kobart_tokenizer()
>>> model = get_kobart_model()
>>> # model = BartModel.from_pretrained("hyunwoongko/kobart")

>>> inputs = kobart_tokenizer(['안녕하세요.'], return_tensors='pt')
>>> model(inputs['input_ids'])
Seq2SeqModelOutput(last_hidden_state=tensor([[[-0.4488, -4.3651,  3.2349,  ...,  5.8916,  4.0497,  3.5468],
         [-0.4096, -4.6106,  2.7189,  ...,  6.1745,  2.9832,  3.0930]]],
       grad_fn=<TransposeBackward0>), past_key_values=None, decoder_hidden_states=None, decoder_attentions=None, cross_attentions=None, encoder_last_hidden_state=tensor([[[ 0.4624, -0.2475,  0.0902,  ...,  0.1127,  0.6529,  0.2203],
         [ 0.4538, -0.2948,  0.2556,  ..., -0.0442,  0.6858,  0.4372]]],
       grad_fn=<TransposeBackward0>), encoder_hidden_states=None, encoder_attentions=None)

For Seq2Seq Training

• seq2seq 학습시에는 아래와 같이 get_kobart_for_conditional_generation()을 이용합니다.
• BartForConditionalGeneration.from_pretrained("hyunwoongko/kobart")와 동일합니다.

>>> from kobart_transformers import get_kobart_for_conditional_generation
>>> # from transformers import BartForConditionalGeneration

>>> model = get_kobart_for_conditional_generation()
>>> # model = BartForConditionalGeneration.from_pretrained("hyunwoongko/kobart")

Bugfix Notes

• version 0.1 : pad 토큰이 설정되지 않은 에러를 해결하였습니다.

from kobart import get_kobart_tokenizer
kobart_tokenizer = get_kobart_tokenizer()
kobart_tokenizer(["한국어", "BART 모델을", "소개합니다."], truncation=True, padding=True)
{
'input_ids': [[28324, 3, 3, 3, 3], [15085, 264, 281, 283, 24224], [15630, 20357, 3, 3, 3]], 
'token_type_ids': [[0, 0, 0, 0, 0], [0, 0, 0, 0, 0], [0, 0, 0, 0, 0]], 
'attention_mask': [[1, 0, 0, 0, 0], [1, 1, 1, 1, 1], [1, 1, 0, 0, 0]]
}

Reference

• SKT KoBart
• Huggingface Transformers